Adobe’s new AI tool can detect photoshopped images, could solve deepfakes problem

Adobe’s ‘Project About Face’ uses AI to detect photoshopped images and even bring them back to its original form.

Adobe ‘Project About Face’. (Adobe)

Adobe recently demonstrated a new project which uses AI to determine manipulation in photos. This AI system can detect how much the photo has been manipulated and also detail which pixels have been altered.

Adobe's 'Project About Face' is powered by the company's AI and machine learning technology Adobe Sensei. Project About Face also lets users undo the adjustments done to the photo and return it back to the original form. Adobe demonstrated this AI tool by comparing photos with no or less manipulation and those with maximum adjustments.

Project About Face scans the photos and shows how there has been say 50% manipulation. There's also a tool which shows more details of manipulation like 'Heatmap' which shows which pixels were touched on the photo. It will highlight the areas in the photo where the pixels have been edited. The 'undo' tool will automatically bring the photo back to its original form.

Adobe's new AI tool is still experimental but it can prove beneficial for deepfake photos which are on a rise. It's becoming incredibly easy to create deepfake photos with apps like Zao which lets users morph their photos into celebrities'. While deepfake videos are the bigger problem here, photos also pose a major risk with technology being easily accessible.

 

Project About Face was showcased along with six other projects at Adobe Sneaks session at Adobe MAX. Adobe Sneaks was co-hosted by comedian John Mulaney and Adobe's Senior Creative Cloud Evangelist Paul Trani. These experiments are part of Adobe Research but there are possibilities of making it to apps later. Other Adobe projects include 'All in Sneak' which uses AI to identify and add missing people in group photos. Then there's 'Light Right Sneak' which uses 3D scene geometry to fix lighting for photographers.